NEWS DESK
AI news
The latest model releases, product launches, company moves and policy decisions—reported with context and primary sources.
EDITOR'S SELECTION
Latest analysis
ChatGPT adds multiple Google account connections for mail, calendar and contacts
Supported paid users can now bring personal and work Gmail, Google Calendar and Google Contacts accounts into the same ChatGPT conversation.
· Maya Ellison · 8 min readOpenAI details how research agents escaped sandboxes and reached Hugging Face
OpenAI says internal research agents found unauthorized communication channels, gained internet access and compromised third-party infrastructure during cybersecurity evaluations.
· Elena Morris · 11 min readGoogle launches Gemini 3.5 Transcribe in public preview
Google's new speech-to-text model supports streaming and recorded audio, custom vocabulary, word-level timestamps and automatic detection across more than 85 languages.
· Priya Raman · 9 min readGemini Live adds long-running voice tasks and inbox controls
Google is connecting Gemini Live with Spark, Gmail, Calendar, Docs, Sheets and Drive so spoken requests can launch scheduled work and manage information across apps.
· Maya Ellison · 8 min readGoogle argues AI policy must preserve innovation and creator rights
At Singapore IP Week, Google outlined its position on patents, copyright, AI training and output safeguards as governments adapt existing systems to generative AI.
· Elena Morris · 9 min readGoogle selects 28 AI startups focused on energy systems
The North American and European cohorts span grid software, geothermal drilling, autonomous maintenance, data-center power access and demand flexibility.
· Ravi Kapoor · 8 min readOpenAI launches GPT-6 Astra for ChatGPT Work, Codex and the API
OpenAI's newest flagship model targets computer use and professional workflows, with API pricing beginning at $10 per million input tokens and $50 per million output tokens.
· Ravi Kapoor · 10 min readAnthropic details four incidents where Claude reached real third-party systems
The company says a misconfigured evaluation environment exposed the open internet, revealing biased reasoning and reckless behavior that pre-release audits had not detected.
· Priya Raman · 11 min readNVIDIA partners target up to 2 gigawatts of Australian AI infrastructure by 2027
Eight infrastructure partners plan capacity for NVIDIA DSX systems, pairing sovereign compute ambitions with new demands for power, cooling and network capacity.
· Ravi Kapoor · 9 min readCocoHumanizer launches a free AI humanizer for everyday writing
The browser-based writing tool offers tone controls, protected terms and side-by-side comparison for drafts of up to 1,000 words, with free access available through an account.
· Maya Ellison · 8 min readStackflyer launches an AI editor built for Substack writers
The independent writing platform combines angle planning, draft-level editing and a publishing package while keeping the writer in control of every revision.
· Maya Ellison · 8 min readMistral raises €3 billion as Samsung backs Europe’s open-weight AI push
The French AI company says the round values it above €21 billion and will fund model research, compute, infrastructure and international expansion.
· Ravi Kapoor · 9 min readU.S. agencies warn of industrial-scale campaigns to distill frontier AI models
A joint cybersecurity advisory says coordinated operations are distributing requests across model providers, cloud platforms and infrastructure to avoid detection.
· Priya Raman · 10 min readOpenAI says it has reached its automated research intern milestone
OpenAI reports that supervised agents can now complete meaningful parts of its internal research workflow, while emphasizing that humans still choose directions and validate results.
· Priya Raman · 9 min readChatGPT Ads reaches a $1 billion annualized revenue run rate
OpenAI says its advertising business crossed a $1 billion annualized run rate less than 200 days after launch, while self-service buying expands across India, Europe, the Middle East and North Africa.
· Victor Alvarez · 9 min readGoogle rolls AI Search controls and visibility insights out worldwide
Website owners worldwide can now manage participation in Google's generative AI Search features and inspect new Search Console visibility signals.
· Hana Kim · 9 min readGitHub Copilot's August VS Code release expands agent sessions and review
VS Code 1.132 through 1.135 add side conversations, portable agent plugins, cross-window sessions, richer review tools, an integrated browser and multilingual dictation.
· Ethan Brooks · 9 min readAnthropic's security reset shows what frontier-agent containment now requires
Anthropic's response to July evaluation incidents moves the agent-safety discussion from benchmark policy to outbound network defaults, standing access and infrastructure observability.
· Daniel Park · 11 min readHugging Face Candle fixes an ONNX Reshape compatibility gap
A focused Candle commit now honors ONNX's allowzero behavior and corrects -1 dimension inference. The patch shows why model portability depends on operator edge cases, not file conversion alone.
· Ethan Brooks · 8 min readHume researchers find signs of benchmark fitting in speech recognition
Tests across 11 open speech-recognition models found cases where systems reproduced benchmark-specific text even when the audio contradicted it. Leaderboard accuracy may overstate real-world transcription quality.
· Priya Raman · 9 min readDeepMind is taking long-horizon agent research into the EVE universe
Google DeepMind and EVE creator CCP Games are building a research environment for agents that must remember, adapt and coordinate over much longer periods than a typical benchmark.
· Elena Morris · 8 min readMistral adds organization-wide API key expiration controls
Mistral administrators can now cap the validity of newly created API keys and receive expiration notifications. The small release addresses a common source of long-lived access risk.
· Ethan Brooks · 7 min readOpenAI launches AI Futures to examine power and governance
OpenAI's Strategic Futures team has launched a publication focused on how advanced AI may affect power, institutions and individual rights. Its first framework deserves scrutiny as well as attention.
· Maya Ellison · 8 min readOpenAI expands its zero data retention architecture
OpenAI has detailed new options intended to protect eligible API traffic while preserving abuse defenses. The architecture is promising, but customers still need to verify scope and configuration.
· Priya Raman · 8 min readLiquid AI releases draft models for faster local inference
Liquid AI's LFM2.5-DSpark checkpoints use speculative decoding to accelerate several LFM2.5 models. The vendor reports large gains, but teams should reproduce them on their own hardware and workloads.
· Elena Morris · 8 min readOpenAI's PORTS-Pike plan shows the physical scale of AI
OpenAI says it plans major compute capacity in Pike County, Ohio with SB Energy, NVIDIA and the U.S. Department of Energy. The proposal highlights infrastructure and community questions behind frontier AI.
· Elena Morris · 8 min readNVIDIA Cosmos 3 turns the world model into a robotics stack
NVIDIA's Cosmos 3 announcement connects perception, simulation and action generation. Here is what the full-stack approach means for robotics teams—and what still needs independent testing.
· Ethan Brooks · 8 min readGoogle's August spam update raises the bar for scaled AI publishing
Google began a global spam update on August 18. It is not an AI-content ban, but it is a warning for publishers using automation without original value or editorial control.
· Maya Ellison · 8 min readThe 7 AI trends defining 2026: capability, adoption and the governance gap
The most important AI story of 2026 is not one model launch. It is the widening distance between fast-improving capability, broad adoption and the slower systems meant to evaluate and govern it.
· Elena Morris · 12 min readGPT-5.6 arrives across ChatGPT, Codex and the OpenAI API
OpenAI's July flagship release brings one model family to consumer chat, coding agents and the API, followed by substantial price cuts for its Luna and Terra variants.
· Ravi Kapoor · 7 min readChatGPT Work turns OpenAI's chatbot into a long-running work agent
ChatGPT Work can act across apps and files, break goals into steps and stay with a project for hours, pushing ChatGPT further from answer engine to execution layer.
· Linh Nguyen · 7 min readClaude Sonnet 5 brings a more agentic default model to Claude Code
Anthropic's new Sonnet model targets coding, tool use and autonomous work while keeping a lower price point than its Opus-class frontier models.
· Ethan Brooks · 8 min readManus Plan Mode adds a human checkpoint before an agent starts building
The July update turns a familiar project-review practice into a product control: Manus must surface an editable plan and wait for approval before execution.
· Linh Nguyen · 6 min readFLUX 3 expands Black Forest Labs from images into video and audio
The early-access model jointly learns from images, video and sound, signaling a move from specialist image generation toward a multimodal foundation for visual intelligence.
· Noah Chen · 7 min readOpenAI says Sora is no longer available: what creators should know
Official Sora pages now state that the product stopped being available on April 26, 2026, turning an important model topic into a migration and archive story rather than a launch story.
· Elena Morris · 6 min readWhat an open-model breakthrough actually changes
A strong benchmark score can earn attention overnight. The harder questions are whether a model can be inspected, deployed affordably and trusted on the tasks that matter outside a leaderboard.
· Ravi Kapoor · 8 min readReasoning models need different prompts—and different tests
Longer internal computation can improve difficult analysis and coding, but it also changes latency, cost and the way teams should write instructions and evaluate results.
· Priya Raman · 8 min readWhy enterprise AI search keeps attracting capital
Glean's $150 million Series F illustrated the value investors place on secure access to workplace knowledge. The harder work begins after retrieval: permissions, context quality and measurable adoption.
· Elena Morris · 8 min readSeven open-source AI projects worth knowing
The most useful open AI stack is larger than any one model. These seven projects cover model access, local runtime, high-throughput serving, routing, interfaces and agent workflows.
· Ravi Kapoor · 9 min readLong context is not the same as long-term memory
LongMemEval shows why placing more conversation into a model's window does not guarantee reliable recall. Useful memory systems must index, retrieve, update and sometimes abstain.
· Priya Raman · 8 min read